> But I care to be able to distinguish between a non-supplied argument and > #f as these are, logically, two different things. > > Maybe I would like to handle a tri-state keyword argument: > true/false/use whatever default value. The default should be the > default, not #f. > > Example: > > (parameterize ((default-case-sensitivity #t)) > (foo) > (foo case-sensitivity: #t) > (foo case-sensitivity: #f)) > > Languages become simpler if they do not lump together things that are > logically different. A tri-state is a classic example of why I would prefer all default values to be #f. Programmers love logic puzzles, and Common Lisp's intricate syntax encourages us to create puzzles that the users of those procedures then have to solve. Probably most CL programmers go through a puzzle-happy phase once they discover lambda lists, macros, the condition system and CLOS. It's fun for them, but less fun for people who have to use their creations :) Those users usually include themselves, once a few months have passed and details forgotten. Emacs is a perfect example of the kind of large, long-lived system that benefits most from keyword arguments. The usage for a typical (not pathological) Emacs subsystem is like this: <https://www.gnu.org/software/emacs/manual/html_node/elisp/Defining-Faces.html>. When you take into consideration the amount of documentation needed for these systems, then as a user, I'm happy if there is nothing sophisticated in those systems that doesn't absolutely have to be there :) The more "idiot-proof" the usage can be, the better. Not only is the usage daunting in a clear-headed, well-rested state, but if you're tired, have little time or other pressures, it gets even more difficult. Emacs deals with case-sensitivity by having a dynamic variable `case-fold-search` (it works just like a Scheme parameter object). If a case-sensitivity argument is needed, I would either go with that, or use symbols or an abstract case-sensitivity dataytype or something. Again, my point is total complexity of large systems. It's difficult to convey using examples, since we can always invent a task that is more convenient using one kind of syntax or another. But in a system big enough that you still feel lost after months, simple universal semantics really pay off. > Scheme is a better language because '() and #f are different objects. After a year of intensive Scheme usage, I'm still not sure which way is better. Coming from CL, the lack of nil-punning took a lot of mental adjustment.
